Read AI Transcription: Speakers, Languages, Uploads and Exports

Read AI transcribes every meeting its assistant joins on Zoom, Microsoft Teams or Google Meet, live during the call and again in a full report afterward. That report carries a transcript split by speaker, and it is usually ready 5 to 15 minutes after the meeting ends. Free covers 5 reports a month (reports shared with you count) for meetings up to 1 hour; paid plans have no report cap. Read AI's help center lists 35 languages, and the product picks the language on its own. Two gaps matter most. Speaker labels disappear in desktop, mobile and uploaded recordings, and Read AI publishes no accuracy rate for its transcription.

How Read AI records and transcribes a meeting

By default the assistant joins from your connected calendar as a visible participant, named read.ai meeting notes, and posts a recording notice in the chat. While the call runs, the live meeting dashboard shows a running transcript and bullet summaries refreshed about every minute, next to talk time and filler words. When the meeting ends, Read builds the report: a summary, chapters, action items, key questions and the full transcript by speaker. It needs at least 50 words of speech. Below that, no report appears.

So does Read AI record meetings? Yes. It captures the call to build the report, but replaying audio or video starts on Enterprise; Free and Pro users get the transcript and the written report only. On Zoom, the assistant now joins only if you have connected your Zoom account and are in the meeting yourself, a rule we cover in our Read AI meeting notes guide. Once a calendar is connected, Read joins every event on it by default, the 7-day trial included, and sends the recap to everyone with access (how to stop it).

Not every capture uses the assistant. Google Meet native recording works without a bot after someone approves a consent pop-up within 10 minutes of the start. The Windows and macOS desktop app records computer audio from any platform, including Webex, GoTo Meeting and RingCentral, which the assistant cannot join; it shows a live transcript but captures no video. The iPhone and Android apps transcribe in-person conversations live, though they cannot record phone calls or FaceTime. In both apps Read is invisible to other people. You must tell participants and collect their consent yourself.

Who said what: speaker labels and where they stop

When the assistant attends, the transcript attributes each line to a named speaker. Hybrid rooms are the first weak spot. Everyone sharing one conference-room microphone shows up as Conference Room – Speaker 1, Speaker 2 and so on, until someone with edit rights renames them. The second weak spot is capture without the assistant. Desktop and mobile recordings carry no speaker attribution at all, and uploaded files get no automatic speaker identification. The transcript arrives as one undivided voice.

The practical rule follows from that. If attribution matters, as it does for interviews, sales calls or board minutes, let the assistant join the call rather than recording locally or uploading a file later. Coaching and engagement metrics follow the same pattern: Read creates them only when it joins the meeting, never for uploads, desktop recordings or mobile recordings. For a transcript that already lacks names, an editor can add them by hand, which works but costs time on a long call. Plan for that gap before the meeting, not after.

Languages and accuracy: what Read AI promises

Read AI supports more than 20 languages, and its help center lists 35 for meeting reports ^[1]. There is no language picker. Read detects the primary spoken language and writes the transcript, summary, chapters and action items in it. That design has two consequences. Meetings that switch between languages are not fully supported, and a meeting held in an unsupported language produces no report at all. Some agents built on transcripts are narrower still: Monday Briefing is English only, End of Week covers English and Spanish, and the Daily Read podcast is English only.

On accuracy, the answer is short. Read AI publishes no accuracy rate. Its Terms of Service, last updated in February 2026, treat AI outputs as informational and say Read AI does not warrant their accuracy, so checking names, figures and commitments falls to you. That matters most for the parts people act on: action items, prices quoted on a sales call, dates promised to a client. Read the transcript against the summary before you forward either one, especially after a call with heavy jargon or several accents.

Fixing a transcript after the meeting

Report owners and editors can edit transcript text and speaker names; viewers cannot. By default internal participants get editor access and external ones get viewer access, and anyone who adds Read to a meeting becomes an editor. A revert option restores the original transcript. Use it carefully. Once a revert is saved, it cannot be undone, so any manual corrections made before it are gone for good.

Prevention beats correction. Paid plans add a custom vocabulary of up to 100 terms for names, products and jargon, but it applies only to new reports, so a term added today does nothing for last week's call. Transcription itself can be switched off for future meetings, and a Workspace admin can disable it for everyone; transcript storage can also be limited or turned off while summaries and action items keep arriving. One blind spot to know: Read does not capture the typed meeting chat, so links and notes pasted there never reach the transcript.

Uploading recordings Read AI did not attend

Uploads cover meetings the assistant missed: an in-person session recorded elsewhere, a Teams call started from Chat, a platform Read cannot join. Read accepts MP4 or M4A files up to 5 GB with at least 50 words of speech, and processing takes at least as long as the recording and up to 24 hours ^[2]. M4A files produce no video. Uploaded reports also skip coaching, Read Score graphs and recommendations, on top of the missing speaker labels.

Each minute of audio or video uses one credit. Pro includes 100 upload minutes a month, Enterprise 200 and Enterprise+ 300; monthly minutes reset at the start of each month and never carry over. Free includes no upload minutes, and neither does the trial. Extra upload credits cost $20 per 500 minutes, about $0.04 a minute, never expire and are spent after the monthly allowance. A failed upload deducts nothing.

The arithmetic decides whether to buy credits or upgrade. By our arithmetic, four 60-minute recordings a month use 240 minutes: 140 beyond Pro's allowance and 40 beyond Enterprise's. One 500-minute pack therefore lasts about three and a half months on Pro, or a little over a year on Enterprise. If minutes are the only gap, credits are cheaper. Three packs cover a full year of the extra 100 monthly minutes Enterprise adds, for less than the yearly price difference between the two plans on either billing term. Upgrade for playback, not for minutes.

Length caps, Free's quota, retention and downloads

The length cap applies to each meeting ^[3]. When a call hits it, Read leaves and builds the report from what it recorded up to then, so a 90-minute workshop on Free yields a transcript of the first hour only. Free's quota is just as literal. Reports shared with you count toward the 5, and from the sixth report of the month Read marks them Out of quota, with no transcript. Upgrading reprocesses reports from the past 30 days and unlocks those, though not reports left Incomplete for too little speech.

Downloads are plain. The recap and the transcript download as .txt files; the recording, its 30-second trailer and the 2-minute highlight reel download as .mp4 only with playback on Enterprise or Enterprise+ ^[4]. Read AI documents no PDF, DOCX or SRT export. Teams that need subtitle or document formats will convert the text themselves. A Workspace admin can also disable downloads for members. Playback and pricing per plan are covered in our Read AI pricing breakdown.

Retention depends on the plan. On Free, Read AI keeps report text for at least 1 year and no longer than 2 years. On paid plans, reports stay until someone deletes them, unless an Enterprise+ retention policy removes reports and Ask Read history after a set number of days, past and future alike. Only the report owner can delete a report for everyone; other people can only remove it from their own list. Deleting the account permanently erases the reports it owns. Export anything you need first.

The bottom line

Read AI's transcription is strongest when its assistant sits in the meeting: speaker names, a live transcript, a full report within minutes and, from Enterprise, the recording itself. Step outside that path and the trade-offs pile up quickly, with no speaker labels in desktop, mobile or uploaded recordings, no published accuracy rate, one detected language per meeting and only .txt and .mp4 downloads. Free works for five short meetings a month, each up to 1 hour. Anyone who transcribes daily, uploads interviews or needs longer calls belongs on a paid plan, chosen by playback needs rather than upload minutes.